Organizing Institutions:
Heinz von Foerster Gesellschaft / Wien
ASC – American Society for Cybernetics
WISDOM – Wiener Institut für
  sozialwissenschaftliche Dokumentation und Methodik
Institut für Zeitgeschichte | Universität Wien
AINS – Austrian Institute for Nonlinear Studies

John Holland

Steering Complex Adaptive Systems

Complex adaptive systems (cas) –ecosystems, markets, and the immune system are examples – are difficult to steer because a cashas no central executive andit is made up of diverseagents that learn and adapt as they interact.

The key components of a cas are boundaries and the signals that interact with the boundaries.  As examples:  Biological cells, and organelles therein,are bounded by semi-permeable membranes which pass selected proteins that serve as signals and resources; markets are defined by diverse groups of traders that issue buy and sell signals; immune system antibodies decompose invading antigens to signal other agents in the immune system; and so on.

For many cas we have large databases, but the databases rarely give direct access to these key components.  Moreover, standard mathematical attempts to analyze available cas databases, say statistical analyses, face additional difficulties because of the conditional interactions between cas agents – the difficulties are much like tryingto use statistics to understand the output of a computer program.  A less traditional option for examining cas centers on extracting mechanisms (operators) and the “building blocks” they manipulate(generators) to produce the signals and boundaries used by the cas.  The result is a finitely generated dynamics (fgd) which leads to laws that help to steer the cas, much like Newton’s laws enable the steering of planetary probes

Because cas signals and boundaries are always assembled from copies of a limited variety of elements (generators)– 4 nucleotides for chromosomal DNA ,20 amino acids for proteins, 32-64 chip level instructions for computer programs, 26 letters for English words, and so on – the cascomponents for afgdcan be defined in terms of strings.  In most cases, relatively small parts of the signals, called tags, determine the routing and processing of signals.  Examples of tags are active sites on enzymes, headers in memoranda, and motifs in DNA.Over time,genetic operators (e.g. mutation and crossover)applied tostring-definedtags, determine theprogressive co-evolution of signals and boundaries.  In this author’s opinion, understanding this co-evolutionary process, is a sine qua non for understanding and steeringcas.